The act of asking a question of an AI, whether it’s a request for a quick recipe or an in-depth analysis, feels almost effortless for users—but behind those neatly packaged answers lies a global infrastructure consuming staggering amounts of energy and water. As artificial intelligence tools—from conversational models like ChatGPT to image generators—become deeply woven into daily life, their unseen environmental consequences are rapidly moving to the forefront of debate among industry insiders, sustainability experts, and the communities hosting vast data centers. Nowhere is this reckoning more acute than in regions like Australia, where the collision of AI progress, energy demand, and water scarcity is creating urgent questions about how sustainable this new digital era can be.
Modern AI relies on what are known as “hyperscale data centers”—giant facilities housing tens of thousands of servers and related equipment. “We’re not just talking about buildings the size of a Bunnings warehouse anymore,” says Gordon Noble, Research Director at the Institute for Sustainable Futures at UTS. Today’s largest data centers can approach a million square feet, equivalent to the size of sprawling campuses, and are proliferating at breakneck speed all over the globe.
Recent estimates place the number of hyperscale data centers worldwide at over 1,100. Each one, by necessity, is a nerve center connected to dense high-speed internet backbones, such as the 12 submarine cables that make Sydney a hotspot for the data center industry. In fact, Australia’s largest city now hosts more than 85 such facilities—a number that continues to rise as international traffic and domestic demand for AI applications grow.
To put this consumption into perspective, data centers already account for about 1.5% of global electricity use, as reported by the International Energy Agency (IEA). Within countries where the data center sector is highly concentrated—such as the United States and Ireland—those numbers can spike considerably, placing significant strain on the local grids. In Ireland, for example, the national energy regulator recently reported that data centers consume nearly 18% of the country’s electricity, challenging both grid stability and the nation’s climate commitments.
In Australia, the future trajectory is even more dramatic. A Morgan Stanley research report, cited by Noble, projects that the roughly 5% share of electricity currently drawn by Australia’s data centers could climb to between 8% and 15% by the end of this decade. If realized, such a shift would force tough decisions about energy allocation, emissions reduction, and the viability of Australia’s renewable energy targets.
Globally, the IEA predicts that by 2030, the demand from data centers (driven in large part by AI) could match the current electricity usage of Japan—the world’s third-largest economy. Tech giants are scrambling to respond; Microsoft, for instance, is reportedly looking at reviving dormant nuclear plants in the US to ensure a stable supply of low-carbon power, including the controversial Three Mile Island facility.
As AI models grow (with GPT-4 and its successors reportedly dwarfing even their already-massive predecessors), the energy curve for both training and inference is trending sharply upwards. Each time you ask a ChatGPT-style tool for help, a small but not insignificant portion of a power plant somewhere is working on your behalf.
“Since 2022, two-thirds of new data centers have been developed in areas exposed to water stress,” Noble warns. This is particularly problematic in regions like Australia, one of the world’s driest continents, where water security already sits at the heart of environmental and social policy.
A report by the Uptime Institute indicates that a typical hyperscale facility can use as much as 25 million liters (over 6 million gallons) of water each month. Globally, research published in the journal Nature found that in 2022, the total water footprint from all data centers was estimated to be in the order of billions of cubic meters annually—figures that could soon rival the residential consumption of mid-sized cities.
For Australia, the irony is acute. Sydney and Melbourne, likely hubs for further data center expansion, currently discharge 97% of their recycled water into the ocean. This points to an opportunity, says Noble: “If we’re smart enough to create a system to funnel surplus recycled water into data center cooling, we could meet their needs without additional environmental stress on rivers and streams.”
However, this would require major infrastructure investments and coordinated planning—not in evidence in Australia at a national level, where digital infrastructure policy remains fragmented between jurisdictions and portfolios.
But when the first round of climate-related financial disclosures started to roll in from Apple, Microsoft, Google, and Amazon, there was a surprise: “The picture that was emerging was a very consistent increase in their energy consumption and emissions,” Noble remarks. Different companies showed varied results depending on whether they owned or outsourced their facilities, but a common thread was clear—increased investment in AI and data centers led to rising emissions, not the decreases many had promised.
According to Microsoft’s 2023 Environmental Sustainability Report, total company emissions (including scope 1, 2, and 3 categories, which capture both operational footprint and supply chain) rose by about 30% between 2020 and 2023, primarily due to new data center construction and increased AI workloads. The trend is similar at other tech majors, despite intense lobbying for carbon credits and power purchase agreements with renewable producers.
The challenge is made more acute when increases in AI-related energy demand “crowd out” grid capacity intended for broader electrification and decarbonization. If solar and battery expansion can’t keep up, “you start to put pressure on how much we can actually invest in more renewables...do we keep coal-fired power stations longer than we need?” Noble asks. Such trade-offs are not unique to Australia; European data center hubs face similar strains, prompting some governments to impose moratoria or strict energy rationing for new facilities.
This dynamic prompts a crucial self-reflection for both policymakers and end users. What is the real environmental cost of every AI-generated insight, email summary, or personalized suggestion?
Australia’s digital economy still operates under a regulatory regime with roots in the days of the postmaster general—outdated and ill-suited for the complexity of today’s interconnected landscape. Until robust frameworks for energy, water, and digital infrastructure converge, local communities may be left to absorb the risks and burdens.
Choosing platforms that commit to renewable energy, supporting open-source and smaller models (which generally consume less power), and advocating for greater transparency in tech supply chains are all meaningful steps. “There’s a lot of benefit to AI if we get this right,” Noble stresses—but it is up to governments, tech companies, and civil society to prevent today’s data-driven revolution from derailing broader environmental goals.
It is now urgent to place energy and water at the heart of digital infrastructure planning, set clear emission targets, prioritize recycled water and renewable energy, and demand data transparency from both government and industry. Only by doing so can the promise of artificial intelligence serve not only the digital needs of today but also the environmental realities of tomorrow. Until such measures are firmly in place, each question posed to an AI will remain part of a much bigger—and as yet unresolved—environmental story.
Source: Australian Broadcasting Corporation Why your AI questions are a power and water drain - ABC listen
The Invisible Engine Room: AI’s Unsung Infrastructure
Modern AI relies on what are known as “hyperscale data centers”—giant facilities housing tens of thousands of servers and related equipment. “We’re not just talking about buildings the size of a Bunnings warehouse anymore,” says Gordon Noble, Research Director at the Institute for Sustainable Futures at UTS. Today’s largest data centers can approach a million square feet, equivalent to the size of sprawling campuses, and are proliferating at breakneck speed all over the globe.Recent estimates place the number of hyperscale data centers worldwide at over 1,100. Each one, by necessity, is a nerve center connected to dense high-speed internet backbones, such as the 12 submarine cables that make Sydney a hotspot for the data center industry. In fact, Australia’s largest city now hosts more than 85 such facilities—a number that continues to rise as international traffic and domestic demand for AI applications grow.
The Power Paradox: From Convenience to Carbon
Most people don’t realize that every query to an AI tool draws on resources orders of magnitude greater than a standard web search. “It can be 10 times more energy-consuming than, for instance, doing a Google search,” Noble explains. This difference comes down to how AI models work: rather than retrieving a single piece of information, they’re drawing on vast pre-trained models gargantuan in both scope and computational requirements.To put this consumption into perspective, data centers already account for about 1.5% of global electricity use, as reported by the International Energy Agency (IEA). Within countries where the data center sector is highly concentrated—such as the United States and Ireland—those numbers can spike considerably, placing significant strain on the local grids. In Ireland, for example, the national energy regulator recently reported that data centers consume nearly 18% of the country’s electricity, challenging both grid stability and the nation’s climate commitments.
In Australia, the future trajectory is even more dramatic. A Morgan Stanley research report, cited by Noble, projects that the roughly 5% share of electricity currently drawn by Australia’s data centers could climb to between 8% and 15% by the end of this decade. If realized, such a shift would force tough decisions about energy allocation, emissions reduction, and the viability of Australia’s renewable energy targets.
Globally, the IEA predicts that by 2030, the demand from data centers (driven in large part by AI) could match the current electricity usage of Japan—the world’s third-largest economy. Tech giants are scrambling to respond; Microsoft, for instance, is reportedly looking at reviving dormant nuclear plants in the US to ensure a stable supply of low-carbon power, including the controversial Three Mile Island facility.
Why Does AI Cost So Much Power?
The essential thing that separates an AI application from a basic search isn’t just the scale—it’s the complexity. Large language models (LLMs), like those behind ChatGPT, are trained on enormous datasets encompassing much of the world’s digital content. Every interaction doesn’t just query a database; it activates a swarm of calculations, referencing billions of parameters, working in tandem to produce context-appropriate, conversational responses. The scale of the “thinking”—thousands of high-powered graphics processors humming simultaneously for every answered question—results in energy draw far greater than traditional IT systems.As AI models grow (with GPT-4 and its successors reportedly dwarfing even their already-massive predecessors), the energy curve for both training and inference is trending sharply upwards. Each time you ask a ChatGPT-style tool for help, a small but not insignificant portion of a power plant somewhere is working on your behalf.
The Quiet Toll of Water
The energy appetite of AI is now fairly well-publicized, but its water usage is emerging as a newer, less understood environmental challenge. Data center hardware runs hot—and efficient cooling is vital to prevent costly meltdowns and ensure reliable operation. Many facilities use water-based cooling methods, tapping into local supplies and cycling billions of liters annually through their systems.“Since 2022, two-thirds of new data centers have been developed in areas exposed to water stress,” Noble warns. This is particularly problematic in regions like Australia, one of the world’s driest continents, where water security already sits at the heart of environmental and social policy.
A report by the Uptime Institute indicates that a typical hyperscale facility can use as much as 25 million liters (over 6 million gallons) of water each month. Globally, research published in the journal Nature found that in 2022, the total water footprint from all data centers was estimated to be in the order of billions of cubic meters annually—figures that could soon rival the residential consumption of mid-sized cities.
For Australia, the irony is acute. Sydney and Melbourne, likely hubs for further data center expansion, currently discharge 97% of their recycled water into the ocean. This points to an opportunity, says Noble: “If we’re smart enough to create a system to funnel surplus recycled water into data center cooling, we could meet their needs without additional environmental stress on rivers and streams.”
However, this would require major infrastructure investments and coordinated planning—not in evidence in Australia at a national level, where digital infrastructure policy remains fragmented between jurisdictions and portfolios.
Emissions: The Undeniable Climb
Perhaps the issue most at odds with the tech world’s green rhetoric is the rise in greenhouse gas emissions tied to AI and its supporting machinery. For years, the industry has espoused ambitious decarbonization plans, with big tech companies joining the RE100 campaign and making public commitments to 100% renewable energy.But when the first round of climate-related financial disclosures started to roll in from Apple, Microsoft, Google, and Amazon, there was a surprise: “The picture that was emerging was a very consistent increase in their energy consumption and emissions,” Noble remarks. Different companies showed varied results depending on whether they owned or outsourced their facilities, but a common thread was clear—increased investment in AI and data centers led to rising emissions, not the decreases many had promised.
According to Microsoft’s 2023 Environmental Sustainability Report, total company emissions (including scope 1, 2, and 3 categories, which capture both operational footprint and supply chain) rose by about 30% between 2020 and 2023, primarily due to new data center construction and increased AI workloads. The trend is similar at other tech majors, despite intense lobbying for carbon credits and power purchase agreements with renewable producers.
Grid Strain and the Prospect of Trade-offs
For countries like Australia, where decarbonization efforts are running up against surging digital demand, this creates a conundrum: can a nation aiming to boost renewables afford to keep scaling its AI sector?The challenge is made more acute when increases in AI-related energy demand “crowd out” grid capacity intended for broader electrification and decarbonization. If solar and battery expansion can’t keep up, “you start to put pressure on how much we can actually invest in more renewables...do we keep coal-fired power stations longer than we need?” Noble asks. Such trade-offs are not unique to Australia; European data center hubs face similar strains, prompting some governments to impose moratoria or strict energy rationing for new facilities.
This dynamic prompts a crucial self-reflection for both policymakers and end users. What is the real environmental cost of every AI-generated insight, email summary, or personalized suggestion?
A National and Global Reckoning: Who Owns the Challenge?
Despite early signals from the market, there is as yet no comprehensive national strategy in Australia for aligning the growth of data centers with sustainable energy and water policy. “What we’re seeing in other jurisdictions, for instance, Singapore, is the development of a green data center roadmap,” Noble points out. In Singapore, strict energy-efficiency guidelines, sustainable cooling requirements, and capped growth for new facilities are now in place to prevent runaway demand.Australia’s digital economy still operates under a regulatory regime with roots in the days of the postmaster general—outdated and ill-suited for the complexity of today’s interconnected landscape. Until robust frameworks for energy, water, and digital infrastructure converge, local communities may be left to absorb the risks and burdens.
Charting a Sustainable Path: Opportunities and Dilemmas
The threats posed by unchecked AI data center growth are significant but not insurmountable. There are concrete opportunities for reform and innovation that, if seized, could create a more balanced digital future:- Recycled Water Integration: There is huge potential in constructing pipelines and distribution systems to direct recycled city water into data centers, sharply reducing freshwater withdrawals.
- Siting Data Centers Strategically: Future facilities can be located in cooler areas, near abundant clean energy, and, ideally, water supplies that are sustainable and not needed for agriculture or residential use.
- Investment in Carbon-Free Energy: Greater use of wind, solar, hydro, and even advanced nuclear power could decouple AI expansion from fossil fuel reliance—but will require capital, planning, and bold targets.
- Green Data Center Certification: Like LEED standards in the building sector, sector-specific sustainability certifications could be developed and mandated, pushing operators toward best practices worldwide.
- Demand Management and Efficiency: AI models and hardware must become more efficient, with regulatory and market incentives pushing for reductions in both energy and water intensity per transaction.
- Digital Transparency: Tech giants must disclose granular data on water and energy usage, emissions, and location impacts—moving beyond the opaque aggregated reports of today.
What Users Can Do
For individual users, these revelations prompt tough questions about digital habits. Every AI prompt, every creative generation, carries a small but real environmental footprint. While end users cannot solve systemic issues alone, their choices can indirectly drive demand for sustainable digital services.Choosing platforms that commit to renewable energy, supporting open-source and smaller models (which generally consume less power), and advocating for greater transparency in tech supply chains are all meaningful steps. “There’s a lot of benefit to AI if we get this right,” Noble stresses—but it is up to governments, tech companies, and civil society to prevent today’s data-driven revolution from derailing broader environmental goals.
Critical Analysis: Balancing Strengths and Risks of AI Infrastructure
Strengths:- Potential for Positive Impact: Advanced AI can solve complex problems in healthcare, science, and sustainability, offering benefits that could, in some cases, outweigh its environmental footprint—if managed wisely.
- Innovation in Sustainability: The rapid growth of the sector is already accelerating research in energy and water efficiency, with some operators leading the way in green construction and renewable sourcing.
- Job Creation and Economic Opportunity: Data center investments create high-value jobs and new business opportunities, particularly in digital infrastructure, cooling technology, and renewable integration.
- Unmanaged Expansion: Left unchecked, runaway energy and water use from data centers could overwhelm grids, exacerbate water scarcity, and undermine climate commitments.
- Environmental Injustice: Facilities are often sited in communities where residents see few benefits but bear the brunt of local environmental impacts, including heat, noise, and water competition.
- Regulatory Lag: Outdated policy frameworks and fragmented jurisdictional oversight may allow short-term industry pressures to trump long-term sustainability and social priorities.
Toward a Sustainable Digital Era
Australia is at a crossroads emblematic of the world: as AI becomes foundational to everyday life, so does the challenge of building a sustainable digital backbone. The path forward is not about demonizing technology, but neither is it about ignoring the high price tag that comes with each effortless AI query.It is now urgent to place energy and water at the heart of digital infrastructure planning, set clear emission targets, prioritize recycled water and renewable energy, and demand data transparency from both government and industry. Only by doing so can the promise of artificial intelligence serve not only the digital needs of today but also the environmental realities of tomorrow. Until such measures are firmly in place, each question posed to an AI will remain part of a much bigger—and as yet unresolved—environmental story.
Source: Australian Broadcasting Corporation Why your AI questions are a power and water drain - ABC listen
